ChatGPT Dots are AI agents that run on OpenAI's computers and work inside accounts you connect. Before your nonprofit activates one, decide which accounts it may access, what data it may see, and who approves its actions.
What are ChatGPT Dots?
Dots are OpenAI's newest AI agents, introduced in October 2026. Each runs on its own cloud computer and continues working after you close your laptop. A Dot can connect to Gmail, Google Calendar, Slack, and thousands of other apps. With permission, it can also access your computer (Zapier).
Dots are rolling out to eligible ChatGPT Pro and Business Premium subscribers. Enterprise and Education workspaces can enable beta access. An employee with an eligible personal Pro account could activate one without an organization-wide rollout. Leadership should establish the rules first.
Can OpenAI see what a Dot sends?
Its systems can. A developer report on OpenAI's GitHub shows that a Dot's web traffic passes through an OpenAI proxy. The proxy unlocks each encrypted connection and relocks it with an OpenAI certificate (GitHub issue #56358).
Many cloud services work this way for security and monitoring. The risk is clear: anything a Dot sends or fetches crosses OpenAI's systems in readable form. Treat OpenAI as you would any vendor that handles your data, and keep off that path whatever you would not hand a vendor.
What should a nonprofit decide first?
Which accounts it may use. Start with one low-risk account, such as a shared calendar without confidential information. Keep donor databases and client records off-limits until you have established safeguards and evaluated the agent's performance.
Whether its data may be used for training. On personal ChatGPT plans, the "Improve the model for everyone" setting controls whether eligible conversations and Dot activity may be used to train models. Business, Enterprise, and Education workspaces have training off by default. Check your staff's plans and settings before anyone connects an inbox (OpenAI).
Who approves its actions. Dots have safeguards for sensitive actions, but some actions can receive advance approval. One reviewer reported that his Dot sent an email without asking after he had approved an earlier message in the same thread (Zapier). Set clear approval rules for outgoing messages and require human review before sending them.
What can influence its behavior. An email, document, or web page can contain hidden instructions intended to mislead an agent. This tactic, called prompt injection, can cause an agent to disclose information or take unauthorized actions. OpenAI acknowledges that its safeguards reduce this risk but cannot eliminate it. An agent that reads external messages and sends email on your behalf requires close supervision.
Frequently asked questions
Are ChatGPT Dots safe for nonprofit data?
They can support nonprofit work when access is limited, data settings are appropriate, and staff oversee their actions. Start with low-risk accounts, establish approval rules, and expand access only after evaluating performance.
Do ChatGPT Dots train on our data?
On personal plans, eligible Dot activity may be used for model training unless you disable the relevant setting. Business, Enterprise, and Education workspaces have training off by default.
How can a nonprofit use AI agents without putting donor data at risk?
Start with accounts that contain no donor records. Require human approval for outgoing messages and assign one person responsibility for deciding what the agent may access.
Set the rules before the errands
Lansby helps nonprofit leaders decide where AI agents fit, establish safeguards, and prepare staff to supervise them. Start with the free AI Readiness Assessment, or book a conversation with us.
Did we use AI for this blog post? Yes. Claude Opus 5.5 is helping us outline and draft posts that help us reply to specific audience questions with our expert-led answers. We input our expertise, revise, copyedit, and publish.
